FOREST CANOPY AND GPS MOVEMENT DATA

<).i.-,

Effect of forest canopy on GPS-based movement data Nicholas J. DeCesare, John R. Squires, and Jay A. Kolbe Abstract The advancing role of Global Positioning System (GPS) technology in ecology has made studies of animal movement possible for larger and more vagile species. A simple field test revealed that lengths of GPS-based movement data were strongly biased (P<0.001) by effects of forest canopy. Global Positioning System error added an average of 27.5% additional length to tracks recorded under high canopy, while adding only 8.5% to opencanopy tracks, thus biasing comparisons of track length or tortuosity among habitat types. Other studies may incur different levels of bias depending on GPS sampling rates. Ninety-nine percent of track errors under high canopy were <7.98 m of the true path; this value can be used to set the scale-threshold at which movements are attributed to error and not biologically interpreted. This bias should be considered before interpreting GPSbased animal movement data. Key words animal movement, forest canopy closure, Global Positioning System, GPS error, GPStracking, snow-tracking, tortuosity

The study of animal movement can be motivated The rise of GPS tracking in collecting animal by a variety of research questions. Movement data movement data has brought with it much need for have been collected to study foraging patterns evaluation of accuracy and bias. Many authors have (Mclntyre and Wiens 1999, Pochron 2001, Nolet documented discrepancies in GPS fix rates and and Mooij 2002, Fauchald andTveraa 2003), identi- location accuracy of wildlife GPS collars caused by fy behavioral differences across groups (With terrain and forest canopy (Moen et al. 1996, D'Eon I994a,b; Bascompte and Vila 1997; Bergman et al. et al. 2002, Di Orio et al. 2003, Frair et al. 2004), 2000), and reveal effects of differing habitat though results are not entirely consistent (Rempel regimes on populations (With 1994b, Ferguson et et al. 1995, Dussault et al. 1999, Bowman et al. al. 1998, Doerr and Doerr 2004, McDonald and St. 2000). These evaluations of GPS are based on indiClair 2004). Many of the field and analytic tech- vidual points. As the link between GPS and analyniques developed in studies of movement have sis of animal movement continues, testing must focused on invertebrate (Kareiva and Shigesada progress to evaluate how these factors affect a 1983, Turchin et al. 1991) or small vertebrate more complex line, or track, of GPS data. (Leman and Freeman 1985, Edwards et al. 2001) We encountered this issue •when overlaying a populations at local scales. However, the advancing data set of lynx (Lynx canadensis) tracks, collected role of Global Positioning System (GPS) technology through snow-tracking, on habitat layers in a in ecology has made such studies possible for larg- Geographic Information System (GIS). Many tracks er and more vagile species. This includes the direct showed differences in "tortuosity" or "sinuosity" deployment of GPS units on study animals (Turchin 1998, McDonald and St. Clair 2004) (Bergman et al. 2000, Biro et al. 2002,Weimerskirch between open-canopy habitats and closed-canopy et al. 2002) and the use of handheld GPS units to forest stands. Movement through openings record movements, as in snow-tracking (Ciucci et appeared linear and direct, while movement al. 2003). through closed-canopy forest was more tortuous Authors' address: United States Department of Agriculture/United States Forest Service, Rocky Mountain Research Station, 800 E. Beckwith, Missoula, MT 59807, USA; e-mail for DeCesare: [email protected].

Wildlife Society Bulletin 2005, 33(3):935-941

Peer referecd

M

llulktin !*

and wandering. Differential tortuosity in animal movements across habitat patches may be indicative of prey density or habitat quality (Zach and Falls 1976, Edwards et al. 2001, Nolet and Mooij 2002, Fortin 2003) or of foraging patterns like the area-restricted (or area-concentrated) search (Walsh 1996, Fortin 2003). However, the patterns in our data also may have been the misleading effect of disparate GPS accuracy across habitat types. It is important to address this issue before interpreting GPS-based wildlife tracking data. We conducted a simple field test to quantify the effect of forest canopy closure on our GPS tracking data by collecting a test data set in the field. These data elucidate the effect of canopy on our GPS tracking, as well as the scale at which this effect is important.

Study area We collected these data in the Clearwater River drainage of the Lolo National Forest, near Seeley Lake, Montana (N 47° 11',W 113° 29'), roughly 50 km northeast of Missoula, Montana. Elevations on the study area ranged from 1,200-1,900 m, and the average slope of data collection sites was 7.5 degrees. Low-elevation forests were dominated by Douglas-fir (Pseudotsuga menziesii), western larch (Larix occidentalis), lodgepole pine (Pinus contorta), and ponderosa pine (Pinus ponderosa), usually as mixed stands. Upper-elevation forests were mostly subalpine fir (Abies lasiocarpa), whitebark pine (Pinus albicaulis), and Engelmann spruce (Picea engelmannif) with lesser components of lodgepole pine, Douglas-fir, and western larch. Subalpine forests were multi-storied and multiaged, often with a dense shrub understory (United States Forest Service 1997).

Methods Data collection During December 2003 and January 2004, we collected a data set of test tracks by using compass bearings to walk straight lines across the landscape. For each line we classified overhead forest canopy closure into 1 of 3 simple categories: open (0-10%), low (11-39%), and high (>40%). We selected areas where classification into these categories was unequivocal and boundaries between categories were clear, such as where frozen lakes or logged units abutted denser forest stands. We attempted to

hold constant travel speeds between categories and collected data during a range of dates and times between 16 December 2003, and 28 January 2004, and between 0830 and 1630 hrs. We collected all data with 4 Trimble GeoExplorer® 3 data logging GPS units (Trimble Navigation Ltd., Sunnyvale, Calif.) and Trimble external antennae mounted at roughly 1.8 m above ground. The units were configured with a 2-second log interval between points, and we differentially corrected data using Trimble GPS Pathfinder® Office software. Base station data came from multiple TRS base stations within 260 km of our study area using 1-second and 5-second logging rates.

Effect of canopy We collected 95 straight-line tracks (wopen = 40, « low =19, «high=36) totaling 19-9 km of GPS length. For each GPS recorded track, we required a paired "true" track to represent the actual distance traveled without GPS error. We initially quantified this distance with a straight line connecting the first and last points of each GPS track. Because we walked only straight lines in collecting GPS tracks, we knew any fine-scale deviation from straight represented error. However, the GPS tracks also showed very gradual bends away from the straight lines, due to difficulties in maintaining a consistent compass bearing over hundreds of meters. These gradual bends were a potentially confounding source of error, and we objectively controlled for this by creating "true" tracks using the Bezier smoothing algorithm in the ET GeoWizards® extension (www.ian-ko.com) for ArcGIS® Desktop 8.3 (Environmental Systems Research Institute, Inc., Redlands, Calif). We found the Bezier algorithm removed all fine-scale GPS scatter and fitted a smoothed line over the actual path (whether straight or gradually bending); thus, we assume this accurately represented the true track. We then used the lengths of the original GPS track and the smoothed, true track to calculate an index (e) of GPS error: e - 1 - (true length/GPS length). Each value of e represents the proportion of a track's original GPS length that is attributable to error. We used one-way univariate analysis of variance (ANOVA) and the type III sums of squares F-test to assess differences in error among categories of

h>rcst cjriufjy .inri GI'S muwmenl cldM • DcCe^are c:l ul.

canopy closure. Given a significant ANOVA result, we used theTukey test (Tukey 1953) as an a posteriori, multiple-comparison test of differences between canopy categories (Zar 1999).

Scale of error

937

recorded in different canopy types, we performed a final test using the maximum PDOP values recorded for each GPS track. We used one-way ANOVA to test if this metric of precision was affected by canopy closure, as a concluding test of the effects of canopy closure on GPS error.

We separated each GPS track into the network of individual GPS points, or vertices, that were sequenResults tially collected in making it. We then used the Nearest Features v. 3.6d (www.jennessent.com/ Effect of canopy Mean error index values (SE) for each canopy catdownloads/NearFeat.zip) extension in ArcView GIS 3.2a to compute the perpendicular distance of each egory were: e open = 0.085 (0.016), elow = 0.154 vertex from our smoothed, true track. We expected (0.033), and ehigh = 0.275 (0.028). Thus, GPS error correlation among error distances within each accounted for an overestimation of track length track and, thus, computed distribution-free toler- from 9-28% depending on canopy closure. ance intervals, containing 99% of the population of Boxplots revealed a positive trend between increaserror distances with 95% confidence (Hahn and ing canopy closure and GPS error (Figure 1), and Meeker 1991). These tolerance intervals represent- ANOVA sums of squares .F-test results suggested ed the range of movement attributable to error canopy was a significant (^292 = 17.78, P< 0.001) within each canopy category, or more importantly, variable in predicting GPS error. Tukey multiple the scale-threshold at which movements could be comparison tests of differences between canopy categories revealed strong differences between biologically interpreted. open and high (%,igh-j>open =0 1 9 0 , P<0.001) and low and high (j>high-j>low=0.121,P=0.008), and no Confounding sources of error = We acknowledged several additional sources of difference between open and low CPiow "Jopen GPS error that had the potential to confound our 0.070, P=0.178). analysis of canopy closure and error. The Trimble GeoExplorer 3 units have standard precision of 1-5 m with differential correction (Trimble Navigation, 0,80" Ltd. 1999) and the smoothing procedures used to generate our "true" tracks have an unknown error. However, we can assert that these sources of error were objective and unbiased across habitat types. Differences in user speed, topography, and hourly satellite availability across habitat types could also confound this analysis. For each track we quanti- Hi fied user speed in meters walked per second, topography as the slope of each track centroid, and £ satellite availability as the number of available satel- C9 0 2 0 lites and the available Position Dilution of Precision (PDOP), a metric of satellite geometry, based on daily satellite almanacs in the Quick Plan operator of Trimble GPS Pathfinder Office software. We then High (MK>«) LOM(11-30«) Op
9'JH

\ritni]

Uniterm Jons

Open (0-10%)

Low (11-39%)

High (>40%)

r

*

30%

Median = 0.139 99'i'i Tolerance Inteival = (0.00,2.53) Range (0.00, 7.33)

Median = 0.214 99% Tolerance Inteival = (0.00,3.29) Range (0.00, 8.45)

Median =0.525 99% Tolerance Inteival = (0.00,7.98) Range (0.00,30.8)

C 20%

o

a. 10% 1 2

3

4

5

Error Distance (m)

1

2

3

4

5

Error Distance (m)

1

2

3

4

5

6

Error Distance (m)

Figure 2. Histograms showing the distributions and variability of error distances associated with each vertex of GPS track data collected under three different categories of canopy closure, Seeley Lake, Montana, 2004. Tolerance intervals are distribution-free intervals containing 99% of error distances with 95% confidence.

Scale of Error

treated as actual movement instead of GPS error. Error distances, measured from each GPS track vertex (n = 9,817), ranged from 0-31 m, and his- Confounding sources of error tograms by canopy type revealed that high canopyOne-way ANOVA showed insignificant differclosure forests were associated with track vertices ences across canopy closure for user speed (F2 80 = of larger and more variable error (Figure 2). Errors 1.461, P=0.238), available satellites (F2t92 = 0.696, P were manifested as oscillations, or "zig-zag," about = 0.501), available PDOP (F292 = 0.077, P=0.926), the true path, causing an increase in track tortuosi- and slope (F2i92 = 1.44,P=0.24l). Thus, we expect that error caused by these sources was relatively ty as canopy increased (Figure 3). Median error distances were small (Mopen = 0.139 m, M low =0.2l4 m,Mhigh = 0.532 m) but increased with canopy. Error distance tolerance intervals indicated with 95% confidence that 99% of errors were within 0.00-2.53 m, 0.00-3.29 m, and 0.007.98 m for open-, low-, and high-canopy forests, respectively. Errors were greatest under high canopy, with 99% of error distances <7.98 m of the true track. This latter value sets the scale-threshold at which movements can no longer be biologically interpreted; thus, in an analysis of actual movement data, only moveFigure 3. Straight-line test tracks (n = 6, displayed) overlayed on a 1995 USGS digital ments that deviate >7.98 orthophoto quadrangle reveal visual differences in GPS error between open (black line) and m from straight should be high (white line) canopy closure, Seeley Lake, Montana, 2004.

Forest canopv nnd GPS movement cljta • DeCesaie et al.

consistent across canopy-closure types and did not confound our analysis. As a concluding test, an ANOVA of the maximum PDOP actually achieved during tracks of each canopy type did reveal differences CF292 = 4.67,P-0.0l2). While available satellite configurations were equal across canopy types, forest canopy did appear to interfere with the satellite pseudoranges actually received by the GPS units.

Discussion Our study demonstrates that GPS error, as caused by forest canopy, has the potential to create bias in movement data. Such habitat-induced bias can lead to spurious measures of movement and type I errors in data analysis. The mean error index («?) under high canopy was 0.275; thus, an average of 27.5% of track length recorded under high canopy was attributable to GPS error. Preliminary inspection of actual track data suggested that this percentage could be larger under the wide range of forest and topographic conditions encountered during actual snow-tracking sessions (J. R. Squires, United States Forest Service - Rocky Mountain Research Station, unpublished data). These data were collected with Trimble GeoExplorer 3 data loggers, and the effect of canopy may differ among GPS manufacturers and units (Walton et al. 2001, Di Orio et al. 2003); we were unable to test differences among GPS units. Global Positioning System error also may vary across seasons (Dussault et al. 1999), ecotypes (Dussault et al. 1999), or topographic conditions (D'Eon et al. 2002), and we encourage caution when extrapolating our results to different study sites. Ideally, we recommend that researchers complete a similar test with project-specific methods and conditions before analyzing GPS tracking data. We used a 2-second log interval between the collection of each GPS point along tracks. This interval affects the resolution with which tracks are recorded and may influence the scale of incurred error. With our data, this error was represented by spikes in the GPS track <31 m away from the true path. Global Positioning System tracking studies of vagile animals using deployed collars and much greater log intervals (>30 minutes) are quantifying movement at a much coarser scale and are unlikely to be affected by this fine level of error. These studies may be more troubled by differing fix rates among habitat types (D'Eon 2OO3,Frair et al. 2004).

939

However, GPS units have been deployed with very frequent log intervals (as low as 1-second, Biro et al. 2002,Weimerskirch et al. 2002), or handheld units have been used to snow-track (Ciucci et al. 2003) or follow (Pochron 2001) animals. Data collected with these frequent intervals may be affected by the canopy-associated error described in this paper. In such cases the effects of canopy should be considered before interpreting any measures of track length or tortuosity. High- (>40%) canopy forests caused greater error than low- (10-39%) and open- (0-10%) canopy forests, whereas low and open did not differ. This suggests that a nonlinear, or asymptotic, relationship may exist between canopy closure and GPS accuracy. Previous studies have generally used linear statistics to evaluate this relationship (Rempel et al. 1995, D'Eon et al. 2002) when there may in fact be some minimum threshold of canopy closure required before GPS accuracy is affected. Global Positioning System depictions of animal movement may require some editing before further interpretation. The first step is to remove the obvious anomalies or "impossible" points often found in GPS data (D'Eon et al. 2002). Manual editing cannot, however, objectively remove the fine-scale oscillations that still obscure the true path when canopy-associated error is present (Figure 3). Smoothing algorithms, such as the Bezier algorithm in the ET GeoWizards extension for ArcGIS Desktop 8.x, offer an objective and repeatable method of removing error. Before smoothing, it is important for researchers to identify the scale at which error is affecting their GPS tracking data. The scale of this oscillating error represents the scale at which movements can no longer be biologically interpreted and should be used to set a minimum scale of movement analysis. In other words, any deviations from a straight line that occur below this minimum scale may be attributable to error rather than truth. In our case the Bezier smoothing algorithm removed the finest-scale oscillations caused by forest canopy and estimated a line that best represented the movement path through the landscape. Any smoothing procedure will result in the loss of information by reducing turning angles at each sampling interval to an artificially smooth, averaged path. However, this smoothed path is a more conservative representation of the movement path, given that deviations at the finest resolution may be fraught with bias. To quantify the scale of GPS error, we calculated

940

II ili/li/f Snarly llutlctni Jims.

«ii]

error distances for each track vertex to generate a distribution of errors among canopy types. These error distances may underestimate actual GPS error slightly because they are calculated as perpendicular distances from the true line, without accounting for the angle of error. For this reason, we used conservative 99% tolerance intervals to identify the scale of error for each canopy category, and the largest of these estimates sets our minimum scale for future movement analysis. Median error distances were <1 m for all canopy categories, and tolerance intervals suggested that most errors were <7.98 m (Figure 2). Much error was within the predicted 1-5-meter precision of GeoExplorer 3 units, but the effects of forest canopy did widen the range of expected error. Though error at this scale might seem negligible, its cumulative effect may insert a considerable bias on estimates of track length and tortuosity. Acknowledgments. Funding for this project was provided by the Bureau of Land Management, Clearwater National Forest, Idaho Department of Transportation, and Region 1 of the United States Department of Agriculture Forest Service. We are grateful for their support. The Missoula Technology Development Center loaned valuable GPS equipment to the project, and we thank field technicians S. Blum, A. Landro, J. Martini, and E. Patton for their assistance in collecting these data. T. Ulizio, G. McDaniel, and R. King provided valuable review and comments.

Wildlife Society Bulletin 30:430-439. Di ORIO.A. P.,R. CALLAS,ANDR.J. SCHAEFER. 2003. Performance of

two GPS telemetry collars under different habitat conditions. Wildlife Society Bulletin 31:372-379. DOERJR.V.A.J.,AND E. D. DOERR. 2004. Fractal analysis can explain individual variation in dispersal search paths. Ecology 85: 1428-1438. DUSSAULT, C , R. COURTOIS, J. P. OUELLET, AND J. HUOT.

1999.

Evaluation of GPS telemetry collar performance for habitat studies in the boreal forest. Wildlife Society Bulletin 27: 965-972. EDWARDS, M.A., G. J. FORBES, AND J. BOWMAN. 2001. Fractal dimen-

sion of ermine, Mustela erminea (Carnivora: Mustelidae), movement patterns as an indicator of resource use. Mammalia 65:220-225. FAUCHALD, R, AND T. TVERAA. 2003. Using first-passage time in analysis of area-restricted search and habitat selection. Ecology 84:282-288. FERGUSON, S. H., M. K. TAYLOR, E. W. BORN, AND F. MESSIER.

1998.

Fractals, sea-ice landscape and spatial patterns of polar bears. Journal of Biogeography 25:1081-1092. FORTIN, D. 2003. Searching behavior and use of sampling information by free-ranging bison (Bos bison}. Behavioral Ecology and Sociobiology 54:194-203. FRAIR, J. L., S. E. NIELSEN, E. H. MERRILL, S. R. LELE, M. S. BOYCE, R. H. M. MUNRO, G. B. STENHHOUSE, AND H. L. BEYER.

2004.

GPS collar bias in habitat selection studies. Applied Ecology 41:201-212. HAHN, G.J., AND W. Q. MEEKER.

Removing

Journal of

1991- Statistical intervals: a guide

for practitioners. John Wiley and Sons, New York, New York, USA. KAREIVA, P., AND N. SHIGESADA.

1983. Analyzing insect movement

as a correlated random walk. Oecologia 56:234-238. LEMAN, C.A.,AND P.W. FREEMAN. 1985. Tracking mammals with flu-

orescent pigments: a n e w technique. Journal of Mammalogy 66:134-136. MCDONALD,W, AND C. C. ST. CLAIR. 2004. Elements that promote highway crossing structure use by small mammals in Banff National Park. Journal of Applied Ecology 41:82-93. MCINTYRE, N. E.,ANDJ.A.WIENS. 1999. Interactions between land-

literature cited BASCOMPTE, J., AND C. VILA.

1997.

Fractals and search paths in

mammals. Landscape Ecology 12:213-221. BERGMAN, C. M.,J.A. SCHAEFER, AND S. N. LurncH. 2000. Caribou movement as a correlated random walk. 364-374. BIRO, D.,T. GUILFORD, G. DELL'OMO, AND H. E LIPP.

Oecologia 123: 2002. How the

viewing of familiar landscapes prior to release allows pigeons to home faster: evidence from GPS tracking. The Journal of Experimental Biology 205:3833-3844. BOWMANJ. L.,C.O. KOCHANNY,S.DEMARAIS,AND B. D. LEOPOLD. 2000.

Evaluation of a GPS collar for white-tailed deer. Wildlife Society Bulletin 28:141-145. Gucci, P., M. MASI, AND L. BOITANI. 2003. Winter habitat and travel route selection by wolves in the northern Apennines, Italy. Ecography 26:223-235. D'EON, R. G. 2003. Effects of a stationary GPS fix-rate bias on habitat selection analyses. Journal of Wildlife Management

MOEN,R.J.PASTOR,Y.COHEN.AND C.C.SCHWARTZ.

terrain.

1996. Effects of

moose movement and habitat use on GPS collar performance. Journal of Wildlife Management 60:659-668. NOLET, B.A.,AND WF. MOOIJ. 2002. Search paths of swans foraging on spatially autocorrelated tubers. Journal of Animal Ecology 71:451-462. POCHRON, S.T. 2001. Can concurrent speed and directness of travel indicate purposeful encounter in the yellow baboons

(Papio bamadryas cynocepbalus) of Ruaha National Park, Tanzania? International Journal of Primatology 22:773-785. REMPEL, R. S.,A. R. RODGERS,AND K. F.ABRAHAM.

1995. Performance

of a GPS animal location system under boreal forest canopy. Journal of Wildlife Management 59:543-551. TRIMBLE NAVIGATION, LTD. 1999. GeoExplorer 3: datasheet and

specifications. California, USA. TUKEY, J. W

67:858-863. D'EON,R. G.,R. SERROUYA, G. SMITH.AND C. O.KOCHANNY. 2002. GPS

radiotelemetry error and bias in mountainous

scape structure and animal behavior: the roles of heterogeneously distributed resources and food deprivation on movement patterns. Landscape Ecology 14:437-447.

Trimble

Navigation, Ltd., Sunnyvale,

1953. The problem of multiple comparisons.

Department of Statistics unpublished report. University, Princeton, New Jersey, USA.

Princeton

Forest canopy and GPS movement data • DeCesare et al. 9 4 1 TURCHIN, E 1998. Quantitative analysis of movement. Sinauer Associates, Inc., Sunderland, Massachusetts, USA. TURCHIN, R, E J. ODENDAAL, AND M. D. RAUSHER. 1991. Quantifying

insect movement in the field. Environmental Entomology 20:955-963. UNITED STATES FOREST SERVICE. 1997. Rice Ridge ecosystem man-

agement area and watershed analysis vegetation report (unpublished). United States Forest Service, Lolo National Forest, Seeley Lake, Montana, USA. WALSH, E D. 1996. Area-restricted search and the scale dependence of patch quality discrimination. Journal of Theoretical Biology 183:351-361. WALTON, L. R., H. D. CLUFF, P. C. PAQUET, AND M. A. RAMSAY. 2001.

Performance of 2 models of satellite collars for wolves. Wildlife Society Bulletin 29:180-186. WEIMERSKIRCH, H., F. BONADONNA, F. BAILLEUL, G. MABILLE, G.

DELL'OMO,AND H. ELIPP. 2002. GPS tracking of foraging albatrosses. Science 295:1259. WITH, K. A. 1994«. Using fractal analysis to assess how species perceive landscape structure. Landscape Ecology 9:25-36. WITH, K. A. 1994ft. Ontogenetic shifts in how grasshoppers interact with landscape structure: an analysis of movement patterns. Functional Ecology 8:477-485. ZACH, R., AND J.B. FALLS. 1976. Ovenbird (Aves:Parulidae) hunting behavior in a patchy environment: an experimental study. Canadian Journal of Zoology 54:1863-1879. ZAR,J. H. 1999. Biostatistical analysis. Fourth edition. Prentice Hall, Upper Saddle River, New Jersey, USA.

Nick DeCesare (above) works with the Rocky Mountain Research Station in Missoula, Montana, studying lynx ecology in northwest Montana. He received an M.S. in wildlife biology from the University of Montana in 2002, studying the movements and resource selection of bighorn sheep in western Montana. In 2000 he received a B.A. in environmental science and a B.M. in classical guitar performance from Northwestern

University in Evanston, Illinois. John Squires (above) is a research wildlife biologist with the Rocky Mountain Research Station located in Missoula. For his M.S. degree, he studied the effects of oil development on prairie falcons in north-central Wyoming, and studied trumpeter swan ecology in the greater Yellowstone area for his Ph.D. at the University of Wyoming. From 1991-1997 he was a resident research biologist for the Rocky Mountain Research Station studying seasonal changes in the habitat-use patterns of northern goshawks nesting in Wyoming. John currently is the team leader for threatened, endangered, and sensitive species research for the Missoula Wildlife Unit. He currently is studying lynx and wolverine ecology in Montana. Jay Kolbe (below) currently is studying

lynx ecology with the United States Forest Service's Rocky Mountain Research Station in Missoula, Montana. He received his M.S. degree from the University of Montana, Missoula, from which he also received his B.S. Associate editor:

White